metadata

language:
  - ur
license: apache-2.0
tags:
  - automatic-speech-recognition
  - robust-speech-event
datasets:
  - mozilla-foundation/common_voice_8_0
metrics:
  - wer
  - cer
model-index:
  - name: wav2vec2-large-xls-r-300m-Urdu
    results:
      - task:
          type: automatic-speech-recognition
          name: Speech Recognition
        dataset:
          type: mozilla-foundation/common_voice_8_0
          name: Common Voice ur
          args: ur
        metrics:
          - type: wer
            value: 61.1
            name: Test WER
          - type: cer
            value: 25.17
            name: Test CER

wav2vec2-large-xls-r-300m-Urdu

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the common_voice dataset. It achieves the following results on the evaluation set:

Loss: 1.5867
Wer: 0.6240
Cer: 0.2579

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 0.0003
train_batch_size: 16
eval_batch_size: 8
seed: 42
gradient_accumulation_steps: 4
total_train_batch_size: 64
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
lr_scheduler_warmup_steps: 200
num_epochs: 100
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
11.1231	8.31	100	3.5275	1.0	1.0
3.4028	16.63	200	3.1164	1.0	1.0
3.1973	24.94	300	2.5590	1.0	1.0
1.3774	33.31	400	1.4729	0.7646	0.3325
0.5427	41.63	500	1.4140	0.6939	0.3001
0.3541	49.94	600	1.4532	0.6580	0.2815
0.26	58.31	700	1.5309	0.6403	0.2726
0.2025	66.63	800	1.5230	0.6310	0.2655
0.171	74.94	900	1.5578	0.6336	0.2632
0.1511	83.31	1000	1.5733	0.6321	0.2635
0.1352	91.63	1100	1.6022	0.6255	0.2608
0.1192	99.94	1200	1.5867	0.6240	0.2579

Framework versions

Transformers 4.17.0.dev0
Pytorch 1.10.2+cu102
Datasets 1.18.2.dev0
Tokenizers 0.11.0